[FFmpeg-user] Advice on using silence removal

Carl Zwanzig cpz at tuunq.com
Sat Aug 21 21:22:52 EEST 2021


What you'd doing is a "noise gate" function.

A few ideas-
Use either the mean volume or 3-6db below the max as the threshold. A more 
complicated version would be to examine the level at maybe half-second 
intervals and use that to determine the levels of the background and spoken 
parts of the clip.

Do the silence removal before the normalize so you're not bringing up the 
noise level along with the speech.

Expand the dynamic range (compand) to push the voice level up and the noise 
level down. Compand can even do a noise-gate, there's a somewhat cryptic 
example in the audio filters doc 
(https://ffmpeg.org/ffmpeg-filters.html#toc-compand).

One thing that will bite is if the recording's automatic gain control is too 
aggressive and gets the background noise at the start/finish to the same 
level as the voice. Not much you can easily do about that but ask for a new 
recording.

Later,

z!



More information about the ffmpeg-user mailing list